PACIFIC JOURNAL OF MATHEMATICS Vol. 41, No.

3, 1972

DIFFERENTIABLE MAPS WITH 0-DIMENSIONAL CRITICAL SET, I

P. T. CHURCH AND J . G. TIMOURIAN

Let /: Mn -> Np be Cn with n p = 0 or 1, let p ^ 2, and let RP-i(f) be the critical set of /. If dim (J?P-i(/)) ^ 0 and dim (/CRP-i(/)) ^ p - 2, then (1.1) at each e Mn, f is locally ? topologically equivalent to one of the following maps: (a) the projection map p: Rn Rp, > (b) :C^C defined by (z) = zd (d = 2, 3, . . .), where C is the complex plane, or
(c) : CxC-> Cxi? defined by T(S,W) = (2S M;, | W | 2 - | Z | 2 ) ,

where w is the complex conjugate of w. In particular, either / is locally topologically equivalent to p at each xeMn, or (n,p)=(2,2) or (4,3). In a sequel the hypothesis on dim f(Rp-i(f)) is eliminated.

For a Cr(r ^ 1) map / : M" - * JVP let i2ff(/) be the set of points x e Mn at which the rank of (the derivative map of) / is at most q. The critical set of / is defined to be Rp^(f) (in case n < p, Rp^{f) = Afn), and according to the Rank Theorem [2, p, 155] at each xe Mn Rp-i(f), f is locally Cr equivalent (1.3) to the map p: Rn-+Rp defined by p(Xi, x2, #*) = (?I, ^2, , ?P). Thus (1.1) is a generalization of the Rank Theorem for n p 0 or 1, and p ^> 2; moreover for n p = 1 it answers a question of Milnor (1,7). Note that while / is only Cn, the maps p} , and are real analytic. Simple examples (1.4) show that "topologically" cannot be replaced by " C w " and that no reasonable classification is possible if p = 1. Propositions more general than (1.1) are also given ((4.7) and (4.9)). Theorem (1.1) was announced in the talks [2] and [20]. For n = p = 2 (1.1) was essentially proved by Stoilow [2, pp. 147 and 148] and for n = p ^ 3 by Church [2, p. 155]. Both [6, p. 72, (1.5)] and [2, p. 159] deal with maps having a small singular set, and [13, 11] discusses maps with isolated critical points. The map is due to N. Kuiper [13, p. 102] and it is topologically equivalent to the cone map of the Hopf fibration : S3 -+ S2 (1.10). Convention 1.2. A symbol such as Mn denotes a separable nmanifold, without boundary unless otherwise specified (except for obvious cases). A manifold with boundary may have empty boundary. The boundary of a space X is denoted by X or dX (in case
615

616

P. T. CHURCH AND J. G. TIMOURIAN

X is a manifold), the interior of X by int X, and the closure of X by X or Cl[X]. The distance between two points is d(x, y), and S(x, ) = {y: d(x, y) < s). The space of real (resp., complex) numbers is denoted by R (resp., C), euclidean w-space by Rn, its origin by 0, the closed ball Cl[S(0, 1)] in Rn by Dn, and the sphere 3D by Sn~\ A map is a continuous function, the restriction of a function / to X is denoted by f\X, and the composition of two functions by gf or gof. Homeomorphism of topological spaces and isomorphism of groups is denoted by e&. The map : X x Y-+ X is projection, and c is used for the identity map on a space.

Given maps : X-+ Fand : U> V, define x:XxU-+Vx Y by (x)(x, u) = ((x), (u)). Define the open cone c(X) as the identification space obtained from X x [0, 1) by identifying X x {0} to a point x*, and let the cone map C(J): C(X) +c(Y) be the map induced by f x c. 1.3. If /: Mn ~-> Np and g: Kn > Lp are Cr maps on C manifolds (r = 0, 1, > then / and g are Cr equivalent if and ) r only if there are C diffeomorphisms a: Mn Kn and : Np Lp such > > that goa = of. The map / at x is locally Cr equivalent to g at u if there are open neighborhoods U of x and F of /(#) such that f\U: U> V is C r equivalent to ^ with a{x) = u. A C diffeomorphism is a homeomorphism, and (locally) topologically equivalent means (locally) C equivalent.
DEFINITION
r

1.4. Suppose that / is any one of p, , or , and let TJ\ R R be a C homeomorphism such that rj fixes the origin, > w TJ I (J? A) is a C difeomorphism, where A is the closure of a sequence of points converging to the origin, and the rank of the Jacobian matrix of rj at points in A is zero. Since the rank of / at any point 1 in A away from the origin is not maximal, f) is not locally C diffeomorphically equivalent at the origin to any of the maps in (1.1). Thus in the statement of (1.1) "topologically" cannot be replaced by n "C ". For examples to show that no reasonable classification is possible if p = 1 take height functions on compact manifolds which have an infinite collection of local maxima.
REMARK
n n

1.5. Given M and N manifolds with (possibly empty) boundary, n ^ p, and a map /: Mn Np, we now define the branch > m m set Bf c M\ Let R + = {xeR : xm ^ 0}, let F = Rn~p or R\~p, and p let G = R or Rl (not respectively). Then x$Bf if and only if / at x is locally topologically equivalent to : F x G+G at (0, 0). OccaDEFINITION

DIFFERENTIABLE MAPS WITH 0-DIMENSIONAL CRITICAL SET, I

617

sionally the notation B(f) is used. In (1.1) Bp = while B and B (and thus Bf) are discrete. With the definition of Bf the Rank Theorem [2, p. 155] for n ^ p becomes:
R A N K THEOREM 1.6. dN
p

Iff: Rp^(f).

is C (r

^l),n^

p, and dM

= , then Bf

Question 1.7. (Milnor [13, p. 100, first problem]). Let f: Rn-^RP be a (real) polynomial map with an isolated critical point at 0. For what dimensions n ^ p ^ 2 do nontrivial examples exist? The topic of [13], except for 11, is certain complex polynomial maps /: CnJrl C with 0 as the only critical point. These maps have > a deep and very interesting structure related to exotic spheres, and their properties led Milnor to ask about real polynomial maps. After posing this question Milnor says "It is not quite clear what 'non-triviaP should mean here. Certainly the projection is a trivial example". One natural definition is: f is nontrivial at x if and only if xe Bf, i.e. / is trivial at x if and only if / at x is locally topologically equivalent to the projection map p. In the complex polynomial case the study of a singularity employs a certain fibration, and analogous fibration exists in the real polynomial case. Milnor formulates his ("tentative") definition of nontrivial [13, p. 97 and p. 100] in terms of this fibration; we omit it here because it is technical. While Milnor's definition appears to be quite different from the definition we have given above, Church and Lamotke have shown [4] that they agree (at least for n 4). With our formulation we can ask Milnor's question in other contexts. Let/: Mn~+Np (n ^ p ^ 2) be continuous, Cr (r = 1, 2, ; or oo), or real analytic. For what dimensions (n, p) can / have a nonempty discrete (or 0-dimensional) branch set Bf, and, up to local topological equivalence, what are the examples? For n p = 0 or 1 Theorem (1.1) answers a Cn version of this question, and a fortiori answers Milnor's question for these dimensions. A continuous version is discussed in (4.9). In sequels [5] the hypothesis on dim /(i2p_1(/)) is removed, and analogous results for n p 2 are proved in both continuous and Cn contexts. At first glance it seems very special to consider only the

618

P. T. CHURCH AND J. G. TIMOURIAN

cases 0 <^ n p ^ 2 in these theorems, but for every (n, p) with n p ^ 4 and p ^ 2, Church and Lamotke constructed [4] a continuous counterexample with isolated branch point. Moreover, for n p = 3 and p ^ 6, the nonexistence of examples depends on the Poincare Conjecture. Thus our restriction 0 ^ n p ^ 2 in these papers is reasonable.
DEFINITION 1.8. A map /: X Y is proper if, for each compact set Kd Y, f~ (K) is compact; / is light if, for each y e Y, dim (f~\y)) ^ 0; and / is monotone if each f~ {y) is connected. It is quasimonotone if, for each connected open set U c Y and component V of f-\U),f{V)= U[23, p. 151]. THEOREM 1.9. (Cheeger and Kister [1, p. 151]; see also [2, p. 170]). If f: Mn Nn is a proper map, n ^ p, and Bf , then f is the > projection map of a fiber bundle.

(While they assume that / is monotone, this hypothesis is not used in their proof. The manifolds may have nonempty boundary.) 1.10. If : S3> S2 is the Hopf fibration, then in (1.1) is topologically equivalent to the cone map c{).
REMARK

Proof. Let S\r) and S2(r) be the spheres about (0, 0) of radius r in C x C and C x R, respectively, and let : C x C - {(0, 0)} C x > R - {(0, 0)} and r : S\r) -> S\r) be the restrictions of . Then , is f [13, p. 102, (11.6)], is proper, and since R2{) = {(0, 0)}, B = (1.6). Thus (1.9) is a bundle map over S2 x (0, oo), so that is topologically equivalent to 1 x c [17, p. 53, (11.4)], and the conclusion results. Outline of the Proof of (1.1) 1.11. For almost all the proof we work in a purely topological context, assuming topological analogs of the hypotheses of (1.1) or less. The lemmas of 2 show that for each x e Mn there is a manifold neighborhood U of x such that the restriction map . If Bg = , then g g = f\U, g: U-> f(U), is proper and Bgf)dU= is a bundle map (1.9); thus, in general, g can be viewed as a bundle map with singularities. In 3 we show that g is open. In 4 we suppose that x is not a point component of g~(g{x)) and deduce (4.5) that g is a bundle map near x, which implies that x Bf. (We use 3 and the 'almost bundle property' of 2 here.) Hence for each xe Bf, x is a point component of f~{f{x))', this is the situation of [6] and that paper yields the desired conclusion. Differentiability is used only as it is used in [6], i.e. to deduce in (1.1) that f(Bf) is 0-dimensional

DIFFERENTIABLE MAPS WITH 0-DIMENSIONAL CRITICAL SET, I

619

and nicely embedded. In several cases lemmas are stated and proved in somewhat greater generality than needed here, for use in [5] and [7]. 2* Extended embeddings*

2.1. A map g: Jn~m x Rm -> Lp~m x Rm is called a layer map if for each t e Rm, g(Jn~m x {}) c Lp~m x {}. (In case g is an embedding it is called an isotopy.) The restriction map
DEFINITION g\(Jn"m x {t}):Jn~m x {} >Lp~m x {}

is denoted by gt, and its branch set by B(gt). Frequently it is convenient to view gt as a map of Jn~m into Lp~m. 2.2. Let 7: J9P (int n " p ) x Dp be disjoint embeddings with i(t) = (a,i(t), t) (i = 1, 2, , ft). 7%ew /^er is cm isotopy h: Dn~p x Dp-+Dn~~p x Z^ stw fc , agrees with the identity map on (Dn~px {0}) U (dDn-p x D p ), ^d h(7i(t)) = (,(0), ).
LEMMA

Proof.

Let X {^(0): i = 1, 2, r.X x Dp

, k}. The 7 define an isotopy

> (int Dn~p) x Dp

by sending (*(0), t) to (^), ), which is readily extended to a neighborhood of X by sending (u, t) to (w ^O) + a^t), t) for w near ^O). The desired ambient isotopy h: Dn~p x Dp > D w ~ p x D2' extending 7 is given by Lees' Neighborhood ^-Isotopy Extension Theorem [12, p. 530].
DEFINITION

let Qaf"{y). for each qeQ.

2.3. Let / : M >N be a map, let ye WaN , and Define the embedding : Q x [y] Mn by X(q, y) = q > An embedding v: Q x W-+Mn extending with Q x W ^-> ik%
2

> iV '

commutative is called an extended embedding of Q over W. For each w e W let vw'.Q-+ f~ {w) be the embedding defined by vw{q) = v(q, w). An extended embedding with Q a single point space is called a cross-section over W. (In case / is a fiber bundle, it is a cross-section in the usual sense.)

620 LEMMA

P. T. CHURCH AND J. G. TIMOURIAN

2.4. Let f: Mn >NP be a map, let n ^ p, let y e Np, and let p c f~ (y) Bf be a compact (n p)-submanifold with bicollared boundary (e.g. let P be a closed bicollar of a bicollared compact (n p l)-submanifold Qaf~(y) Bf). (a) Then there is an open p-cell neighborhood D of y and an extended embedding : P x D Mn such that Bf imag = (in the > example let v = \Q x D). (b) Given v and Q as in (a) (or Q = ), 3 > 0, and a family ^ of components of f~\y) Q such that L = \j{J: Je ^ \ has compact closure, L L c Q, and either (i) ^ is finite or (ii) each diam J < 3/3, then there is a p-cell neighborhood E of y in D such that for each Je ^ the component K of f~(E)-\m&g v containing J has (i) d(x, J ) < /3 for each x e K, resp. (ii) diam K < . Proof of (a). We will assume that the reader has read the proofs of the Theorem and the Remark in [1] and has them at hand. Let Mn, Np, and y be denoted by W, Y, and yQ9 and let the (n ^-manifold P' be the union of P and a bicollar of its boundary. For each xeP, there are closed neighborhoods U of y0 in Y and V of x in W, and a homeomorphism h: B(2) x U-+ V such that foh is the projection map onto U (since x$Bf). In fact it is possible to choose the closed neighborhood U and a collection of such embeddings {h: j = 1, 2, , k} on 5(2) x ?7 such that P c UiU i n t ^i(5(l)) and Z " 1 ^ ) ^(B(2)) c int P\ For yeU define , = f~\y) - / (so P c P ' c M,o) and hy)j: B(2) -> My by ^,,-(0 = h3{y,t). The proofs of [1] now yield the embedding required for (a). Proof of (b). Let W e TkP be compact with L a int TF; we may suppose that < d(L, bdy W). Let P be a closed bicollar of Q (P = if Q = 0) and : P x D > M % be as in (a); we may suppose that Z > is sufficiently small and P is a sufficiently small bicollar of Q that d(x, Q) < 3/3 for each a? e imag // (remember that Q = //(Q, y)). Let L' be L int P (since L L a Q, U is compact). Each component of L (resp. of f~\D) imag v) contains one and only one component Kr of L' (resp., of f~\D) - int imag ). Let /r c Np (r = 1, 2, . be closed ) p-cells such that Er+1 c Er and r-^r = M Then it suffices to prove that (*) there exists an integer r such that, for each Je ^ the component H of f~(Er) int imag containing J int P has d(x, J int P) < 3/3 for each xeH. Suppose the contrary. Then there are components Jr of ', components iJ r of f~\Er) int imag with J r c Hr, yr e Hr with d(Jr, yr) ^ 3/3, and paths r c i,. joining some point xr e Jr to yr. We may suppose that r c TF, that xr a?0, and that yr i/0. Thus x0 eL', = > >

DIFFERENTIABLE MAPS WITH 0-DIMENSIONAL CRITICAL SET, I

1

621

limsup/V is connected [23, p. 14, (9.1)], and diam ^ <?/3 Now > is contained in some component J' of Z In case (i) there are only a finite number of components of Z/, and since they are compact, all f but a finite number of the Jr are in fact J ; since d(J\ y0) ^ <5/3, a contradiction results. In case (ii), since d i a m J ' < 3/3 and d i a m ^ <?/3, a contradiction results also. 2.5. Let f: M ~^N be a map with 0 ^ n p and n p p 4 or 5, let ye N , let dim (Bf f~\y)) ^ 0, let X cz Bf f~ (y) be compact, and let > 0. Then there is a compact (n p l)-manip fold Q (or ), an open p-cell neighborhood D of y in N , and an n extended embedding v: Q x D-+ M such that each component K of f~\D) imag v meeting X has diam K < , their union has compact closure, and imag v Bf = .
LEMMA
n n

Proof. Let , T c f~\y) be compact with X c int T\ T c int T (interior relative to f~(y)), and let > 0 be less than both d(X, bdy T7') and d(T\ bdy ). We may suppose that e < . Let X' = ' 5/, and let ?7, = {x e : d(a?, Xr) ^ 1/fe} ( = 1, 2, . .). We will first prove that (1) for k sufficiently large, each component of Uk has diameter less than . Suppose the contrary. Then there are a subsequence {m(k)}, components k of Um(k), and points xk, yke k with d(xk, yk) ^ . We may suppose that xk -* x0 and yk yQ. > Thus J = limsup k is connected [23, p. 14, (9.1)], and since d(x0, y0) ^ e, diam ^ . If fc is fixed, then for each j ^ k, 3 c J7m(i) c Um{k), so that c Um{k). Thus c &'mfc) = X, contradicting the fact that X is totally disconnected (since X d Bf f] f"\y)). Thus (1) is true. If a generalized continuum fails to be locally connected, it fails at (at least) a subcontinuum of points [23, p. 19, (12.3)]. Since f~\y) Bf is an (n p)-manifold (or ) and Bf f] f~{y) is totally disconnected, each component of f~{y) is locally connected. Hence (2) if U is open in , then each component of U is open in . Let k be the number given by (1), and (3) let V be a component of Uk meeting X. Then diam V< , so that V a T. Let bdy V refer to the boundary of V in the component v of f~{y) containing V, and let x ebdy V. Since 7 c Uk, d(x, Xr) g 1/fc; suppose d(x, Xf) = < I/A:. The component W of {^ e Av: d(x, u) < 1/fc a} containing x is open in v (by (2)), WaUk, and thus W c i n t F (relative to r), contradicting the choice of x; thus (4) d(x; X) = I/A for each x e bdyF. Let z be the closure of the union of bdy V for V satisfying (3) (actually the union is closed). For each xe A, d(x, X') I/A, and since

622

P. T. CHURCH AND J. G. TIMOURIAN

J c f and T Bf = X', f] Bf = . Let L be the (n - p)-manifold int T Bf (possibly empty). Siebenmann and Kirby have shown (see [15, p. 949] that a topological manifold with dimension not 4 or 5 has a handle decomposition, so in particular there are compact (n p)manifolds with boundary {Lj} (j = 1, 2, , k or j = 1, 2, such ) that Lj c int Lj+1 and L = [J3 Lj. Since is compact, there is a j such that c int Lj. Since dL3 is collared, there is a compact (w n n n p)-manifold P with boundary (or ^) such that c int P , P c int L, , n % n (and so P cz L), and 3P is bicollared in L. Let Q = dP . Since each component Y of / (p) Q meeting X is contained in some V satisfying (3), diam Y < . Let and v be the extended embeddings given by (2.4a and b); the conclusion results. 2.6. Let f: Mn -+NP be a map with 0 ^ n - p and n 4,5, let x e Mn, and let dim (Bf /"1(/(^))) ^ 0. Then there is a connected (not neccessarly compact) manifold Kn c Mn with boundary such that xeintKn( = Kn - diP), e closure Kn of Kn in Mn is compact, there is an open p-cell D c Np with f{Kn) c D, and the restriction map g: Kn D is proper with Bg dKn = . >
COROLLARY
-1

For example, let f:R2+R be projection on the first factor, let x = (0, 0), and let K2 = ( - 1 , 1) x [-1, 1]. Proof. We may suppose that x e Bf. Apply (2.5) where X = {x}, and let Kn be the closure in f"\D) of K. Thus Kn is a manifold with boundary, and dKn c imagv (which may be empty). Let Y be a compact subset of the open 2-cell D. Since bdy Kn (bdy taken relative to Mn) is the disjoint union of dKn and a subset of /""^(bdyi)), flf-^y) = / ^ ( ) Kn = / ^ ( ) i H (closure in Mw), and so is compact; thus # is proper. 3* Open maps* 3.1. For /: Mn Np and a 6 M", let (a ) = /(c) ? be the component of f~{f{x)) containing x. If, for every neighborhood U of x, f(x) eint/(C7), then / is open at x.
DEFINITION

3.2. Let f: Mn > Np be a map with n ^ p, and let x e Mn with dim {f~\f{x)) Bf) ^ 0 and (x) {x}. Then f is open at x.
LEMMA

Proof. Let U be an open neighborhood of x in Mn. For n = p, f~\y) Bf is discrete, so the hypotheses cannot be satisfied. Thus n > p, and there is z e ((x) U) - Bf, so that f{x) = f{z) e int f(U).

DIFFERENTIABLE MAPS WITH 0-DIMENSIONAL CRITICAL SET, I

623

Since U is arbitrary, / is open at x.

3.3. Let f: Mn >NP be a map, and let xeMn with (x) = {x}. Then there is an open p-cell D about f(x) such that, if K is the component of f~{D) containing x, f: K * D is a proper map.
LEMMA

Proof. Use the proof of [6, p. 74, (1.14)]. 3.4. Let f: Mn-+Np be a map with n^p and dimif^iy) Bf) ^ 0 for every y e Np, and let Ef be the set of points at which f fails to be open. (a) Then either Ef = (so that f is open) or dim f(Ef) ^ p 1. (b) In particular, if dim f(Bf) ^ p 2, then f is open.
LEMMA

Proof. Let x e Ef. By (3.2) we may as well assume that (1) (x) = {x}; let g: K+D be the proper map given by (3.3). (In case n p 4,5, we could use (2.6) instead.) Then Eg = Ef K. We observe that (2) if VczK is open, then int g{V) , i.e. d i m ^ F ) p [11, p. 46]. If V<- Bg, the conclusion is immediate. If F c Bg, let Ud V be a closed w-cell. The map g\ U: U-> g(U) is light (from the dimension hypothesis), so that [11, p. 91, VI. 7] dim^(C7) ^ n. Since n ^ p, dim g(V) = dim g(U) = n = p. Now suppose that g{K) D and dim g(Eg) ^ p 2. Since g is proper, g(K) is closed, so that D g(K) is a nonempty open subset of D. By (2) int g{K) . Thus D - bdy #(if) is not connected, and hence [11, p. 48] dim bdy g(K) = p 1. There is a 2 e i with 0(2) e (bdyg(i)) g(Eg), and since is open at z, g(z) e int g{K), so a conr tradiction results. Thus (3) dim g{Eg) ^ p 2 implies It is immediate from (2) that (4) dim g(Eg) ^ p 2 implies g~\g{Eg))\ - K. From (1), (3), and (4) if dim g(Eg) ^ p 2, then # satisfies the hypotheses of [23, p. 149, (7.81)] at x, and so g(Eg) locally separates at a;. A contradiction of [11, p. 48] results, and hence dim g(Eg) ^ p 1, yielding conclusion (a). Since Ef c Bf, conclusion (b) follows from (a). 4* O-regular maps* 4.1. Let X and Y be metric spaces, let / : X> Ybe

DEFINITION

624

P. T. CHURCH AND J. G. TIMOURIAN

open, and let xeX. The map / is 0-regular at x if, for every > 0, there is a > 0 such that if ye Y and u, veS(x, ) f~\y), then there is an arc from u to v in S(x, ) n f~\y). If / is proper, onto, and 0-regular at each xe X, then / is 0-regular. 4.2. If Mn and Np are manifolds with boundary, f: Mn> is a map, and x e Mn Bf, then f is 0-regular at x.
LEMMA

The proof is immediate from (1.5).

LEMMA 4.3. Let X be a separable metric space, let A a X be closed, let X A = Mn and Np be manifolds with boundary, let n> p, let f: X> Np be proper, open, and onto, and let B = B(f\Mn) U A (1.5). Suppose that f(B) is nowhere dense, dim (B f~{y)) ^ 0 for each yeNp, and f"(z) is connected for each zeNp f(B). (a) Then f~\y) is path connected for each y e Np. (b) // / is 0-regular, F ^ Sl or [0, 1], and f~\z) ^ F for each p zeN f(B), then f is a bundle map with fiber F.

Proof. Each y e Np has a compact neighborhood V, and f~\ V) is compact. There are zke V f(B) with zk >y, and f~\zk) f~{y) [23, p. 10, and p. 130, (4.32)], so that f~{y) is connected [23, p. 14]. Since f~~{y) is locally connected except possibly at most 0-dimensional set B f~{y), it is locally connected [23, p.19, (12.3)], and so path connected [23, p. 38, (5.2)]. Under the hypotheses of (b) f~{zk) > f~{y) 0-regularly [21, p. 482], and thus ([21, p. 484, Theorem 2] and [22, p. 341, Theorem 5.1]) f-\y)**F. By [8, p. 115, Theorem 7] f\f~\V) is bundle map, and (b) results. This lemma is used in (4.5) and [7], and its considerable generality is required for the latter applications. 4.4. Given /: M -> N , the singular set Af (see [6]) is defined as follows: x e Mn Af if and only if there are open neighborhoods U of (x) and V of f(x) such that f\U: U->V is topologically equivalent to the projection map : V x (x) * V. Thus (1.5) Bf c Af. See (4.6).
DEFINITION
n p

4.5. Let f: Mn Np be a proper map with n^ p, 3Mn > possibly nonempty, Bf c int Mn, dim f(Bf) ^ p 2, and dim {f~\y) p Bf) ^ 0 for each y e N . (a) If (x) {x} (see (3.1)), then there are open neighborhoods U of (x) and V of f(x) such that f\U:U>V is a proper, open, monotone (onto) map.
LEMMA

DIFFERENTIABLE MAPS WITH 0-DIMENSIONAL CRITICAL SET, I

625

Afnu= ).

(b) (c)

If, in addition, //, in addition,

n p 4 , 5 , then f \ U is O-regular. n = p + 1, then f \ U is a bundle map

n n p

(thus

(d) If (x) {x} for every x e M and M and N are connected, then f = oy where is monotone and is a k-to-one covering map.

Proof. By (3.4) / is open. The hypotheses that (x) {X} and dim (f~\f{x)) Bf) ^ 0 imply that n > p. lp^l, then Bf = , and the conclusions follow from (4.2), [18, p. 63, (2.3)], and [19, p. 661, (2.1)]. Thus we may suppose that n > p ^ 2. Let x e (x) Bf. There is a cross section at x over an open p -1 p-cell V c N . Let U be the component of / ( V) containing (x) and let g: U-> V be the restriction of /. Since dim {g~\y) (Bg) 0, each dimg-'^^n-p; thus by [11, p. 91, Theorem VI 7] dim(fl n 2, so that U' = U g~(g(Bg)) is path connected. Let a: Uf V ^(.ff) be the restriction of #; since o: is a proper > map with Ba = , a can be factored a = 7/5, where /3 is a monotone map and 7 is a covering map [18, p. 63, (2.3)]. Now \(V g{Bg)) is a global cross-section of 7, so that [16, p. 77 (6) and (7)] 7 is a homeomorphism. Thus a is monotone, and by [18, p. 64, (2.5); the proof is still valid for dMn and Bf c int Mn] g is monotone onto. Conclusion (a) results. Now suppose that (x) {X} for every x e Mn. Since / is open, and thus quasi-monotone, there is a natural number k such that each f~(y) has at most k components, and for y e Np f(Bf), f~(y) has exactly k components [18, p. 64, (2.5)]. Let yeNp, and let A (i = 1,2, ,h^k) be the components of f~{y). Let Ui and F* be as given by (a) for i9 and let V be a p-cell neighborhood of y such that VafXiVi and the /^ are in distinct components of f~(V). Since / is quasi-monotone, each component of f~ {V) meets some if so there are exactly h components Wi9 where T^c W<. Since f\ W. Wi~^ V is monotone, for each z e V f(Bf), f~{z) has h components; thus h = k. From [18, p. 63, (2.1)] (d) follows. For xe U Bg, g is 0-regular at x (4.2). Thus, to prove (b) it suffices to prove that g (or equivalently /) is 0-regular at each x e Bg. For xe Bg and > 0, let K be as given in (2.5) for g and X {x}, and let : K D be the restriction of #. Thus c S(x, e) c 7. Since > # has a global cross-section, g{u) {U} for every ue U; thus /\(w) ^ {u}, and by (d) = o, where is monotone and is a covering map. Since D is simply connected, f is a homeomorphism, so that is monotone.

626

P. T. CHURCH AND J. G. TIMOURIAN

By (4.3) (a) r\y) is path connected for each yeD. such that S(x, <) c int K. Since ? S(x, ) f~\y) c r\y) c S(x, e) Z" 1 ^) , / is 0-regular at x. Conclusion (b) results.

Choose > 0

If n = p + 1, then #| (C7- g~(g{Bg))) is bundle map (1.9) with fiber a compact, connected 1-manifold F, i.e., E p& S1 or [0, 1], and conclusion (c) follows from (4.3(b)). 4.6. Let f: Mn -+ Np be a proper map with (x) c Bf (3.1) for every xeBf. Then Bf = Af (4.4).
LEMMA

Proof. If x Af, then x Bf. If x $ Bf, then from the hypothesis (x) Bf = . There is an open p-cell neighborhood V of f(x) such that the component U of f~\V) containing (x) is disjoint from Bf. Since f\U: U-^V is also proper, it is a bundle map (1.9), and since V is contractible, the bundle is trivial. Thus x$Af. 4.7. Let f: Mn Np be Cn with n = p or p + 1, Bf , dim /(By) ^ p - 2, wd dim {f~{y) l / ^ 0 /or c/^ 7 e Np. Then / dim / = 2p n 2 and there is a closed subset X c Bf such that dim X < 2p n 2 wd, /or ec/^ xe Bf X, f at x is locally topologically equivalent to the layer map
PROPOSITION

c() x c: D2{n-p+1)

x R2*-n-z _> Dn~p+2 x i?^- % - 2

(see (1.2)), where c() = (see (1.1)) if n = p, and c() = , i.e. (1.10) the cone map of the Hopf fibration : S3 + S 2 , if n = p + 1. Proof. Let X be the set of all xe Bf such that / at x is not locally equivalent to c(); then X is closed. Let xeBf, and let i be the neighborhood of x and g: K>D be the proper map given by (2.6). If n = p, then for each yeD, g~\y) Bf is discrete and dim (g~\y) -B/) ^ 0, so that g is light, i.e. each (u) = {u}. If n = p + 1 it follows from (4.5(c)) that for each u G Bg, (u) = {u}. There is an open p-cell neighborhood U (Z D of f(x) sufficiently small that the component V of f"\U) containing x has V c int K. Since K is compact, the restriction map h: V+U is proper, so that (4.6) Bh = Ah. Since dim f(Bf) ^ p 2 and Bf , p :> 2. In case n = p,h is light, and since dim (JBA) ^ p 2, the Jacobian determinant of h is (locally) nonnegative or nonpositive [3, p. 94, (2.3) and p. 98, (1.7)].

DIFFERENTIABLE MAPS WITH 0-DIMENSIONAL CRITICAL SET, I

627

Xcz Ahf] h~ (Yh), dim (V X) < 2p n - 2; since a e ; was arbitrary

and V is a neighborhood of x, dim X < 2p n 2.
REMARK 4.8. Theorem (1.1) is a Corollary of (4.7). Rank Theorem (1.6).)

In both cases by [6, p. 83, (4.1)] there is a closed set Yh c h(Ah) such that dim Yh < dim h(Ah) and, for each xe Ah h~{Yh), h at x is locally topologically equivalent to c() x c. Thus dim h(Ah) 2p n 2 and dim Ah ^ 2p n 2. Since & | Ah is light, dim A ^ dim fe(AA) [11, p. 91, Theorem VI 7], so that dim Ah = 2p - n - 2, and dim (A c / r 1 ^ ) ) ^ dim Yh ^ 2p - n - 2. Since Ah = Bh and V

(Use the

The next result (4.9) is a topological analog of (1.1). If / : Mp+1 > Np is continuous with p ^ 2 and Bf discrete, then / satisfies the hypotheses of (4.9); in this case the result was proved by Timourian [19]. 4.9. Let f: Mn >NP be a map with n = p or p + 1 and p^2, let dim Bf ^ 0, and let dim f{Bf) ^ 0 . Ifp^S suppose in addition that for each y e f(Bf) and neighborhood W of y9 there is an open p-cell U such that y e U czW and U f(Bf) is simply connected. Then at each x e Mn, f is locally topologically equivalent to one of the maps p, , or of (1.1).
PROPOSITION

Proof. By the first two paragraphs of the proof of (4.7), for each xe Bf there is a neighborhood V of x such that the restriction h: V> U of / is proper and Bh = Ah. By [6, p. 75, (2.3)] at each ueBh,h at u is locally topologically equivalent to or to c(), where is the Hopf map, i.e. to or by (1.10). 4.10. Let f: Jn~m x Rm -* Lp~m x Rm be a Cn~p+1 layer map with n p = 0, 1, or 2, and dim (Bf f~\y, t)) ^ 0 for each m m t e R . Then Bf = Cl[\Jt{B{ft)}: t e R ].
PROPOSITION

By Sard's Theorem (e.g. [2, p. 156]) dim (ft(R^m^(f(t))) ^pm - 1, and by the Rank Theorem (1.6) (Lp~m x {*}) - f{Bf) is dense in Lp~m x {t} for each t e Rp~m. Our proof uses only this last staten p+1 ment, rather than C ~ . Proof. If (x, s) Bf, then there is a layer embedding : ( D ^ x p m m n m m p m w n p D ~ ) x D ->J ~ x # witn (?, s) e int imag, D - c Z/~ , T: D ~ x p m m p m m ( ~ x D )->L ~ x i? projection, and /o = . For each (v, t) e n p int imag , \t: D ~ x D p ~ m Jn~m is an embedding with v e int imag , > and fto\t = t. Thus (v, t) J5(/,), so that (a?, s)

628

P. T. CHURCH AND J. G. TIMOURIAN

Suppose that (x, s) $ Cl[\Jt B{ft)], but (x, s) e Bf. Choose rj > 0 such that S((x, s), rj) Cl[\Jt B(ft)] = , and let K be the set given by (2.5) for /, X {(x, s)}, and = , over (we may suppose) D = U x V, where 7 and V are open (p m) and m-cells, respectively. Let K be the closure of K in / ^ ( t f x V), and let / : J? -> E x F and gt: K n (J % - m x {}) 7 x {} be restrictions of / . Each is a proper map. > Since B(gt) = , each ^ is a bundle map (1.9): call its fiber Ft. For ue V there exists y with (y, u) e (U x {u}) g(Bg). Choose open fc-cell and m-cell A neighborhoods of y and u, respectively, such that ( x A) flr(B^)= ^. Since g\g~\ x J) is a bundle map (1.9), Ft is independent of t for teA. Since % is arbitrary and V is connected, Ft is independent of t for e 7, i.e., g~(y, t) is independent of y and . By (3.2) g is open, and thus [23, p. 152, (8.1) and (8.11)] quasimonotone. Since the number of components of g~\y, t) is independent of y and t, g = fo^, where ^ is monotone and is a covering map [18, p. 63, (2.1)]. Since U x V is simply connected, f is a homeomorphism, so that g is monotone, i.e., Ft is connected. In case n p = 0 each gt is a homeomorphism, so that the open map g is also one-toone and ontothus g is a homeomorphism, contradicting the choice of (x, s). In case n p = 1 or 2 to obtain a contradiction it suffices [10, p. 527, Theorem B and p. 530, Corollary 2] to prove that g is 0-regular. Given (z, u) e Bg and > 0, let c g~{g{zy u)) dK be a closed (n - p)-cell with (s, %) e int T, dT S, = , and dim < . Let Mn = K, Q = dT, and let v: Q x D * i be the extended embedding given by (2.4 (a)); we may suppose (2.4 (b)) that the component Xof g~(D) imag v containing int T has dim X < and X 3i = 0. Since each g~ {y, t) is a compact connected (w p) manifold with nonempty boundary (n - p = 1 or 2) and v(Q x {t}) ^ dT^ Sn~p~, it follows from the cohomology sequence with compact supports of this pair that 9~~\V> 0 v(Q x ^}) has either one or two components. Since g~{y, t) X contains any component it meets, there are two and g~{y, t) X is one of them. Choose S > 0 such that S((z, u), ) c X. Then for every (y, t) e U x V,

so that g is 0-regular at (2;, u). (4.2), ^ is 0-regular.

Since ^ is 0-regular at each

X&Bf

While (4.10) is not used in this paper, [7] refers to (4.10) and (4.11), they will be used elsewhere, and the proof of (4.10) fits natu-

DIFFERENTIABLE MAPS WITH 0-DIMENSIONAL CRITICAL SET, I

629

rally into this section.

For these reasons, they are given here.

REMARK 4.11. The hypothesis dim (Bf f'\y, t)) ^ 0 is (surprisingly) essential. There is a proper layer map / : ([ 2, 2] x R) x R> R x R such that each ft is topologically equivalent to the projection map, / 0 is the projection map, and Bf = ([ 1, 1] x R) x {0}. For example each f~\Q, t) (t 0) might be the union of the three subsets of ( [ - 2 , 2] x R) x {t} denned by: (1) x = sin t^y for \y\ ^ 3r/2, (2) y = St/2 for - 2 ^ x ^ - 1 and (3) 7 = 3/2 for 1 ^ a; ^ 2. Open maps similar to this have / been defined in [14, p. 9] and [9, p. 341].

REFERENCES
1. J. Cheeger and J. M. Kister, Counting topological manifolds, Topology, 9 (1970), 149-151. 2. P. T. Church, Differentiable monotone mappings and open mappings, pp. 145-183 in The Proceedings of the First Conference on Monotone Mappings and Open Mappings, edited by L. F. McAuley, Oct. 8-11, 1970, SUNY at Binghamton, Binghamton, N. Y., 1971. 3. , Differentiable open maps on manifolds, Trans. Amer. Math. Soc, 109 (1963), 87-100. 4. P. T. Church and K. Lamotke, in preparation. 5. P. T. Church, and J. G. Timourian, Differentiable maps with 0-dimensional critical set, II and III (mimeographed). 6. P. T. Church and J. G. Timourian, Fiber bundles with singularities, J. Math. Mech., 18 (1968), 71-90. 7. P. T. Church, and J. G. Timourian, Real analytic open maps (mimeographed). 8. E. Dyer and M. E. Hamstron, Completely regular mappings, Fund. Math., 4 5 (1957), 103-118. 9. W. C. Fox, The critical points of peano interior functions defined on 2-manifolds, Trans. Amer. Math. Soc, 8 3 (1956), 338-370. 10. M. E. Hamstrom and E. Dyer, Regular mappings and the space of homeomorphisms on a 2-manifold, Duke Math. J., 2 5 (1958), 521-532. 11. W. Hurewicz and H. Wallman, Dimension Theory, 2nd edition, Princeton University Press, Princeton 1948. 12. J. A. Lees, Immersions and surgeries of topological manifolds, Bull. Amer. Math. Soc, 7 5 (1969), 529-534. 13. J. Milnor, Singular Points of Complex Hyper surf aces, Annals of Math. Studies No. 61, Princeton Univ. Press, Princeton, N. J. 1968. 14. W. D. Nathan, Open Mappings on Manifolds, Ph. D. dissertation, Syracuse University, Syracuse, 1968. 15. R. Schultz, Some recent results on topological manifolds, Amer. Math. Monthly, 7 8 (1971), 941-952. 16. E. H. Spanier, Algebraic Topology, McGraw-Hill, New York 1966. 17. N. Steenrod, The Topology of Fiber Bundles, Princeton University Press, Princeton 1951. 18. J. G. Timourian, Fiber bundles with discrete singular set, J. Math. Mech., 8 (1968), 61-70. 19. -. Maps with discrete branch sets between manifolds of codimension one, Canad. J. Math., 23 (1969), 660-668.

630

P. T. CHURCH AND J. G. TIMOURIAN

20. J. G. Timourian, Maps with 0-dimensonal critical set, in The Proceedings of Liverpool Singularities Symposium I, edited by C. T. C. Wall, Lecture Notes in Mathematics 192, Springer-Verlag, Berlin, 1971. 21. P. A. White, Regular convergence of manifolds with boundary, Proc. Amer. Math. Soc, 4 (1953), 482-485. 22. , Some characterizations of generalized manifolds with boundary, Canad. J. Math., 4 (1954), 329-342. 23. G. T. Why burn, Analytic Topology, 2nd edition, Amer. Math., Society, Providence 1963. Received March 13, 1970 and in revised form January 17, 1972. Work of the first author supported by NSF Grant GP-6871; that of the second author by NSF Grant GPand NRC Grant % A7357.
SYRACUSE AND UNIVERSITY OF ALBERTA UNIVERSITY, UNIVERSITY OF TENNESSEE,